Research On Location Fingerprinting Algorithm For Indoor Position

With the development of information technology and the popularity of mobile terminals,people are increasingly demanding for indoor positioning services.How to quickly and accurately obtain the indoor positioning information of the mobile terminal has become the focus of research in indoor position.The indoor positioning technology based on WLAN is one of the important research areas of indoor position.It has the advantages of low cost and easy expansion,and it can effectively solve the problems that the traditional positioning technology is easily obstructed by buildings and the positioning cost is high.At the same time,the location fingerprinting algorithm for indoor position is less dependent on devices and more convenient to implement.It has become a key research area in WLAN positioning technology.The location fingerprinting algorithm based on the Extreme Learning Machine(ELM)has the problems of high cost for collecting labeled training data,poor adaptability to the environment and instability in positioning performance.This paper analyzes and optimizes the problems existing in the traditional location fingerprinting algorithms.The main work includes the following aspects.(1)Focusing on the issue that traditional location fingerprinting algorithm based on ELM collecting labeled training data cost too much,a semi-supervised location fingerprinting algorithm based on manifold regularization is proposed.Firstly,based on the manifold hypothesis,the laplacian operator is constructed according to the similarity between the little labeled data and the large number of unlabeled data;Then,the manifold structure in the fingerprinting signal is obtained under the framework of manifold regularization,and it is used as a penalty term to constrain the geometry of the output model;Finally,combined with the extreme learning machine algorithm,the hidden layer node parameters are generated by the stochastic feature mapping,and the hidden layer output weight matrix is solved to construct the non-linear mapping between the high-dimensional fingerprinting signal and low-dimensional position coordinates.So,the proposed algorithm can reduce the demand for labeled training data.(2)Considering the poor adaptive ability of traditional fingerprint positioning algorithms,an incremental fingerprint positioning algorithm based on semi-supervised extreme learning machine is proposed.First of all,the semi-supervised extreme learning machine based on manifold regularization is used to train the labeled data and the unlabeled data together,and an initial positioning estimation model is established;Then,the iterative recursive model parameter is derived by the block matrix algorithm.When the new half-labeled data set is input,it avoids training for the entire data set repeatedly;At last,the appropriate penalty weight is allocated for new training data set,and the Newton's iterative method is introduced to calculate the final model parameters.The model has better performance for timeliness management.(3)To solve the problem that the hidden layer node parameters of ELM fingerprint positioning algorithm generated randomly,reduce the stability in positioning performance,an ELM positioning algorithm based on parallel chaos optimization is proposed.At first,the traditional chaos optimization algorithm is parallelized,so that each optimization variable is independently parallel-mapped by multiple chaotic variables,which overcomes the instability of the optimization effect brought by the randomness of the chaotic variable;Next,it is used to optimize the hidden layer node parameters of the extreme learning machine.A stable positioning model is constructed.